[flang] Preprocessor substitution in bind(C) names

[bcornille]

Issue:

flang does not substitute preprocessor tokens in BIND(C) names. This reproducer contains continuation lines in the same spirit as #78797. This appears to be a workaround in order to get token pasting inside character/string literals. Open to discussing whether this should be intended behavior, but it is used extensively in the a customer codebase and we'd need to find an alternative solution.

Reproducer:

main.F90

program main
  implicit none
#define TEMP_LETTER b
#define VAL c
  integer, parameter :: a_&
    &b_c = 1
  write(*,*) a_b_c
contains
subroutine test_&
&TEMP_LETTER&
&_wrapper_&
&VAL&
& (input) &
  bind(C, name="test_&
    &TEMP_LETTER&
    &_&
    &VAL&
    &_f")
  use iso_c_binding
  integer :: input
end subroutine test_&
&TEMP_LETTER&
&_wrapper_&
&VAL
end program main

preprocessor output

flang-new -E main.F90

      program main
      implicit none


      integer, parameter :: a_b_c = 1

      write(*,*) a_b_c
      contains
      subroutine test_b_wrapper_c(input) bind(C, name="test_TEMP_LETTER_&
     &VAL_f")








      use iso_c_binding
      integer :: input
      end subroutine test_b_wrapper_c



      end program main

gfortran behavior

gfortran -E main.F90

program main
  implicit none


  integer, parameter :: a_&
    &b_c = 1
  write(*,*) a_b_c
contains
subroutine test_&
&b&
&_wrapper_&
&c&
& (input) &
  bind(C, name="test_&
    &b&
    &_&
    &c&
    &_f")
  use iso_c_binding
  integer :: input
end subroutine test_&
&b&
&_wrapper_&
&c
end program main

[klausler]

Why not use stringizing and //?

[klausler]

program main
  implicit none
#define STR2(x) #x
#define STR(x) STR2(x)
#define TEMP_LETTER b
#define VAL c
  integer, parameter :: a_&
    &b_c = 1
  write(*,*) a_b_c
contains
subroutine test_&
&TEMP_LETTER&
&_wrapper_&
&VAL&
& (input) &
  bind(C, name="test_&
    &"//STR(TEMP_LETTER)//"&
    &_&
    &"//STR(VAL)//"&
    &_f")
  use iso_c_binding
  integer :: input
end subroutine test_&
&TEMP_LETTER&
&_wrapper_&
&VAL
end program main

[llvmbot]

@llvm/issue-subscribers-flang-frontend

Author: Brian Cornille (bcornille)

### Issue: flang does not substitute preprocessor tokens in BIND(C) names. This reproducer contains continuation lines in the same spirit as #78797. This appears to be a workaround in order to get token pasting inside character/string literals. Open to discussing whether this should be intended behavior, but it is used extensively in the a customer codebase and we'd need to find an alternative solution. ### Reproducer: #### main.F90 ```Fortran program main implicit none #define TEMP_LETTER b #define VAL c integer, parameter :: a_& &b_c = 1 write(*,*) a_b_c contains subroutine test_& &TEMP_LETTER& &_wrapper_& &VAL& & (input) & bind(C, name="test_& &TEMP_LETTER& &_& &VAL& &_f") use iso_c_binding integer :: input end subroutine test_& &TEMP_LETTER& &_wrapper_& &VAL end program main ``` #### preprocessor output `flang-new -E main.F90` ```Fortran program main implicit none

  integer, parameter :: a_b_c = 1

  write(*,*) a_b_c
  contains
  subroutine test_b_wrapper_c(input) bind(C, name="test_TEMP_LETTER_&
 &VAL_f")








  use iso_c_binding
  integer :: input
  end subroutine test_b_wrapper_c



  end program main

#### gfortran behavior
`gfortran -E main.F90`

program main
implicit none

integer, parameter :: a_&
&b_c = 1
write(,) a_b_c
contains
subroutine test_&
&b&
&wrapper&
&c&
& (input) &
bind(C, name="test_&
&b&
&_&
&c&
&f")
use iso_c_binding
integer :: input
end subroutine test&
&b&
&wrapper&
&c
end program main

</details>

[bcornille]

I can bring this back to the offending codebase, but am I correct in inferring that you would prefer not to support this pattern?

[klausler]

Preprocessors have not been replacing macros in string literals since 1989. Not going to do it here.

[bcornille]

gfortran appears to fail on the provided example.

>> gfortran -E main_stringize.F90
# 1 "main_stringize.F90"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "main_stringize.F90"
program main
  implicit none




  integer, parameter :: a_&
    &b_c = 1
  write(*,*) a_b_c
contains
subroutine test_&
&b&
&_wrapper_&
&c&
& (input) &
  bind(C, name="test_&
    &"//STR(TEMP_LETTER)//"&
    &_&
    &"//STR(VAL)//"&
    &_f")
  use iso_c_binding
  integer :: input
end subroutine test_&
&b&
&_wrapper_&
&c
end program main

Maybe it thinks that those lines are string literals. Guess that would be a bug in gfortran.

[bcornille]

Apparently the preprocessor as invoked by gfortran doesn't stringize.
https://stackoverflow.com/questions/31649691/stringify-macro-with-gnu-gfortran
This presents a challenge for migrating the given pattern.

[klausler]

I have a solution that is portable to both gfortran and flang-new, but it's pretty ugly. I'll think about this some more.

[klausler]

module m
  implicit none
#define TEMP_LETTER b
#define VAL c
#ifdef __GFORTRAN__
#define Q "
#define STR(x) x
#else
#define Q
#define STR2(x) #x
#define STR(x) STR2(x)
#endif
  integer, parameter :: a_&
    &b_c = 1
contains
subroutine test_&
&TEMP_LETTER&
&_wrapper_&
&VAL&
& (input) &
  bind(C, name="test_"//&
&Q&
&STR(TEMP_LETTER)&
&Q//"_"//&
&Q&
&STR(VAL)&
&Q//"_f")
  use iso_c_binding
  integer :: input
end subroutine test_&
&TEMP_LETTER&
&_wrapper_&
&VAL
end

[klausler]

#96987